Reinforcement learning-based energy management for hybrid electric vehicles: A comprehensive up-to-date review on methods, challenges, and research gaps

Abstract

Reinforcement learning is widely used for control applications and has also been successfully implemented for efficient energy management within hybrid electric vehicles. Reinforcement learning algorithms offer various advantages, including fast convergence, broad applicability, stability, and robustness, particularly with the integration of deep and transfer learning. This paper provides a comprehensive understanding of reinforcement learning principles and a critical review of various reinforcement learning methods, states, actions, and rewards used to optimize the energy management performance of hybrid electric vehicles. Furthermore, the advantages and limitations of these algorithms are also discussed. This review reveals that deep reinforcement learning techniques show superior performance in handling complex energy management tasks thanks to their ability to learn from high-dimensional state spaces. Nevertheless, their implementation faces notable obstacles, including computational complexity and generalization across diverse driving conditions. Finally, key research directions for future work and challenges are highlighted in the domain of reinforcement-learning-based hybrid electric vehicle energy management.